REBA: Robotic Emotional-Based Awareness

As autonomous systems are entrusted with tasks of increased scope and complexity, their behaviour is increasingly dependent on the influence and direction of an expanding number of external agents. Reconciling internal goals with tasks originating from a disparate, conflicting, and perhaps, contradictory set of external goals is a challenging undertaking. Autonomous agents should execute external directives, however, finding the appropriate compromises between these directives and their internal beliefs. We present REBA, a biologically-inspired model that prioritises self-preservation through introspectively derived autonomic action. REBA employs emotions and homeostasis to maintain an internal state of equilibrium and issue corrective autonomic actions during periods of imbalance. Through analytical experiments across several scenarios from the autonomous driving domain, we demonstrate that REBA is effective in satisfying system safety requirements and can find suitable trade-offs in the importance allocated to external directives and internal beliefs.